Thin-film lithium niobate (TFLN), characterized by its low optical loss, remarkable optical nonlinearity and electro-optic properties, serves as an outstanding platform for integrated photonics. However, the existing processing techniques introduce potential damage, thereby impeding the development of high-performance optical devices. The specific sources of damage caused by the processing steps remain unclear. In this paper, we demonstrate that the high-energy electron beam utilized in the Electron Beam Lithography (EBL) process introduces material damage and thus significantly increase the optical loss in TFLN. We further illustrate that the introduced damage can be effectively repaired through post-process annealing, enabling the realization of microring resonators with intrinsic Q-factor (Qint) as high as 3.93 x 106 on X-cut TFLN.

A compact and broadband polarization beam splitter (PBS) is proposed based on an asymmetric directional coupler consisting of a hybrid plasmonic waveguide and a ridge waveguide on lithium-niobate-on-insulator. Due to the surface plasma polariton (SPP) effect, the phase-matching condition is satisfied for transverse electric polarization, whereas the transverse magnetic polarization has a significant phase mismatch by choosing reasonable widths of both waveguides. A short (similar to 28 mu m long) PBS is designed while the gap width is chosen to be 200 nm to make it easy to fabricate. Numerical simulations show that the designed PBS has a broad bandwidth (>130 nm) for an extinction ratio of >15 dB and a large fabrication tolerance for the variation of the waveguide width (over +/- 50 nm). (c) 2023 Society of Photo-Optical Instrumentation Engineers (SPIE)

A polarization beam splitter (PBS) is proposed based on an asymmetric directional coupler consisting of a silicon nitride assisted waveguide and two ridge waveguides on a lithium niobate on insulator. Due to the silicon nitride buffer layer, the phase-matching condition is satisfied for TE polarization, whereas TM polarization has a significant phase mismatch by choosing reasonable widths of three waveguides. A 112.2-mu m-long PBS is designed, and the gap width is chosen to be 300 nm to make it easy to fabricate. Numerical simulations show that the designed PBS has a broad bandwidth (> 80 nm) for an extinction ratio of > 15 dB and an ultralow insertion loss (< 0.35 dB). (c) 2022 Society of Photo-Optical Instrumentation Engineers (SPIE) [DOI: 10.1117/1.OE.61.12.125102]

Airy beam and Airy vortex beam with intriguing characteristics have been extensively studied in linear optics, while their dynamic nonlinear conversion has rarely been explored and remains challenging. Here, through the collaboration of binary mask engineered nonlinear photonic crystal and space‐variant liquid‐crystal geometric phase element, second‐harmonic Airy beam and Airy vortex beam can be generated in a switchable manner. Electrical‐controlled loading of orbital angular momentum (OAM) into fundamental wave unlocks the flexibility of nonlinear process, which is well investigated in theory and further verified by experiment. Wavelength transformation, OAM conversion, and propagation trajectory inflection are realized simultaneously. This offers a convenient and universal strategy for dynamic nonlinear generation and shows great potentials in optical manipulation and information processing.

We experimentally demonstrate the use of a periodically poled LiNbO3 (PPLN) crystal to produce an ultra-long sub-diffraction optical needle via second-harmonic generation. When we input a radially polarized fundamental wave from a femtosecond laser, a sub-diffraction beam size of 0.45λSH and an ultra-long depth of focus (DOF) of 55λSH are experimentally realized, where λSH is the second harmonic wavelength. The lateral size of the second harmonic optical needle is reduced by a factor of 2 compared to the case using the fundamental wave. The compact experimental configuration can realize wavelength conversion and wave-front shaping simultaneously in a single PPLN crystal. In addition, the ultra-long DOF is potentially useful in observing thick samples. The unique characteristics of our second harmonic optical needle open the door for practical applications in super-resolution imaging and optical manipulation.

A continuously deformative space possesses trivial or nontrivial topological characteristics depending on the associated homotopy groups associated with spaces describing the physical processes. Moreover, the interaction of spatial warping and structural symmetry always presents fantastic phenomena, especially in the systems with unique symmetrical properties such as quasicrystals. Here, we propose a quasi-periodic structure (QPS) with topological defects. The analytical expression of the corresponding Fourier spectrum is derived, which reflects the combined effects of topological structure and quasi-translational symmetry. Light-matter interaction therein brings unusual diffraction characteristics with exotic evolution of orbital angular momentum (OAM). Long-range correlation of QPS resulted in multi-fractal and pairwise distribution of optical singularities. A general conservation law of OAM is revealed. A liquid crystal photopatterned QPS is fabricated to demonstrate the above characteristics. Dynamic reconfigurable manipulation of optical singularities is achieved. Our approach offers the opportunity to manipulate OAM with multiple degrees of freedom, which has promising applications in multi-channel quantum information processing and high-dimensional quantum state generation.

Stochastic gradient descent(SGD) is the fundamental sequential method in training large scale machine learning models. To accelerate the training process, researchers proposed to use the asynchronous stochastic gradient descent (A-SGD) method in model learning. However, due to the stale information when updating parameters, A-SGD converges more slowly than SGD in the same iteration number. Moreover, A-SGD often converges to a high loss value and results in lower model accuracy. In this paper, we propose a novel algorithm called Trend-Smooth which can be adapted to the asynchronous parallel environment to overcome the above problems. Specifically, Trend-Smooth makes use of the parameter trend during the training process to shrink the learning rate of some dimensions where the gradients' directions are opposite to the trends of parameters. Experiments on MNIST and CIFAR-10 datasets confirm that Trend-Smooth can accelerate the convergence speed in asynchronous training process. The test accuracy that Trend-Smooth achieves is shown to be higher than other asynchronous parallel baseline methods, and is very close to the SGD method. Moreover, Trend-Smooth can also be combined with other adaptive learning rate methods(like Momentum, RMSProp and Adam) in the asynchronous parallel environment to promote their performance.

As a continuous transition between the Hermite-Gaussian and Laguerre-Gaussian beams, the Ince-Gaussian beams form a family of exact orthogonal solutions of the free-space paraxial wave equation in elliptic coordinates. Ince-Gaussian beams have multiple transverse mode patterns, which make them unique in terms of application in the fields of bioengineering, particle manipulation, and quantum entanglement. Here, based on binary nonlinear computer-generated holograms with a domain structure (realized via electric field poling at room temperature), we generate a second-harmonic Ince-Gaussian beam pumped with a fundamental Gaussian beam. In this process, the transverse part of the phase-matching condition is satisfied, which is called the Raman–Nath-type nonlinear diffraction. Both frequency conversion and beam shaping can be realized simultaneously, thereby offering the advantage of integration of both functions into a single device.

One of the most significant bottleneck in training large scale machine learning models on parameter server (PS) is the communication overhead, because it needs to frequently exchange the model gradients between the workers and servers during the training iterations. Gradient quantization has been proposed as an effective approach to reducing the communication volume. One key issue in gradient quantization is setting the number of bits for quantizing the gradients. Small number of bits can significantly reduce the communication overhead while hurts the gradient accuracies, and vise versa. An ideal quantization method would dynamically balance the communication overhead and model accuracy, through adjusting the number bits according to the knowledge learned from the immediate past training iterations. Existing methods, however, quantize the gradients either with fixed number of bits, or with predefined heuristic rules. In this paper we propose a novel adaptive quantization method within the framework of reinforcement learning. The method, referred to as MQGrad, formalizes the selection of quantization bits as actions in a Markov decision process (MDP) where the MDP states records the information collected from the past optimization iterations (e.g., the sequence of the loss function values). During the training iterations of a machine learning algorithm, MQGrad continuously updates the MDP state according to the changes of the loss function. Based on the information, MDP learns to select the optimal actions (number of bits) to quantize the gradients. Experimental results based on a benchmark dataset showed that MQGrad can accelerate the learning of a large scale deep neural network while keeping its prediction accuracies.

One of the most significant bottleneck in training large scale machine learning models on parameter server (PS) is the communication overhead, because it needs to frequently exchange the model gradients between the workers and servers during the training iterations. Gradient quantization has been proposed as an effective approach to reducing the communication volume. One key issue in gradient quantization is setting the number of bits for quantizing the gradients. Small number of bits can significantly reduce the communication overhead while hurts the gradient accuracies, and vise versa. An ideal quantization method would dynamically balance the communication overhead and model accuracy, through adjusting the number bits according to the knowledge learned from the immediate past training iterations. Existing methods, however, quantize the gradients either with fixed number of bits, or with predefined heuristic rules. In this paper we propose a novel adaptive quantization method within the framework of reinforcement learning. The method, referred to as MQGrad, formalizes the selection of quantization bits as actions in a Markov decision process (MDP) where the MDP states records the information collected from the past optimization iterations (e.g., the sequence of the loss function values). During the training iterations of a machine learning algorithm, MQGrad continuously updates the MDP state according to the changes of the loss function. Based on the information, MDP learns to select the optimal actions (number of bits) to quantize the gradients. Experimental results based on a benchmark dataset showed that MQGrad can accelerate the learning of a large scale deep neural network while keeping its prediction accuracies.

In this paper, we experimentally demonstrate the second harmonic generation of long-range surface plasmon polaritons via quasi-phase matching in lithium niobate. After depositing a 9/13 nm thick Au film on periodically poled lithium niobate, TiO2 of about 2.3 μm in thickness is evaporated on the sample as a refractive-index-matching material. This dielectric (periodically poled lithium niobate)-metal(Au)-dielectric(TiO2) sandwich structure can support the transmission of long-range surface plasmon polaritons through it. By designing a moderate ferroelectric domain period of periodically poled lithium niobate, the phase mismatch between the fundamental wave and second harmonic wave of the long-range surface plasmon polaritons can be compensated and a second harmonic wave can be generated effectively. This can be used to provide integrated plasmonic devices with attractive applications in quantum and classic information processing.

Lithium niobate (LiNbO3) nonlinear photonic crystals have been widely used in nonlinear wave front modulation for the π-phase difference between the second-harmonic (SH) waves from positive and negative domains. Aided by optimizing algorithms, we can control the diffraction patterns of SH vortex beams by designing nonlinear fork gratings with supercell structures in LiNbO3 crystals. In experiment, we demonstrate that the diffraction energies can be equally distributed in each order or concentrated on certain orders, in agreement with theoretical designs. Potential applications include controllable vortex beam generation, nondestructive parallel vortex detection, and so on.

We develop an approach to fabricate arbitrary ferroelectric domain patterns on lithium niobate film (30-50 μm thick) by applying a structured external field at room temperature. The fabricating method can be operated easily to reach 1 μm linewidth resolution. The ferroelectric domain inversion is stable and uniform. Nonlinear diffraction is generated when the fundamental wave pumps to film. Various nonlinear wavefronts are obtained such as the frequency converted optical vortex beam. A nonlinear holographic concept is proposed to explain the physical phenomena and guide the corresponding domain design. The applications in optical field manipulation and novel photonic states generation are discussed.

Geometric phases have attracted considerable attention in recent years, due to their capability of arbitrary beam shaping in a most efficient and compact way, while traditional geometric phases are usually limited to handling single-structured beams and lack the capability of parallel manipulation. Here, we propose a digitalized geometric phase enabling parallel optical spin and orbital angular momentum encoding. The concept is demonstrated in inhomogeneous anisotropic media by imprinting a particularly designed binary phase into a space-variant geometric phase. We theoretically analyze its spin orbit interaction of light and experimentally created higher-order Poincare sphere beam lattices, the order number and symmetry of which can be flexibly manipulated. Special lattices of cylindrical vector beams and orbital angular momentum modes with square and hexagonal symmetry are presented. This work discloses a new insight in programming geometric phases for tailoring the optical field and inspires various photonics applications.

Based on nonlinear holography, we propose a 2D spirally poled LiNbO3 nonlinear photonic crystal that generates orbital angular momentum (OAM) states of second-harmonic (SH) waves. In this crystal, the generated SH waves from positive and negative domains have a π phase difference, which is used to compose a nonlinear Fresnel zone plate for an experimental demonstration of generating SH OAM states at the designed focusing spot. In addition, the crystal can be used to detect input OAM states of fundamental beams without significantly disturbing their wave fronts.

The band structure of surface plasmon polaritons (SPPs) on the Ag surface in the presence of gratings and SPP-based surface-enhanced Raman scattering (SERS) are investigated theoretically and experimentally. The SPP bandgap position can be tuned by geometric parameters. The SPP band edge dominates the SERS behavior. The template stripping process is introduced to reduce SPP propagation losses, improving SERS sensitivity by ∼40. Apart from flexibility and a moderate SERS enhancement factor of the order of 105–106, the SPP band structure is highly reproducible with a relative standard deviation of 10.9%. Our results open opportunities for SPP band structures to serve as SERS substrates.

Ferroelectric domain inversion and its effect on the stability of lithium niobate thin films on insulator (LNOI) are experimentally characterized. Two sets of specimens with different thicknesses varying from submicron to microns are selected. For micron thick samples (~28 um), domain structures are achieved by pulsed electric field poling with electrodes patterned via photolithography. No domain structure deterioration has been observed for a month as inspected using polarizing optical microscopy and etching. As for submicron (540 nm) films, large-area domain inversion is realized by scanning a biased conductive tip in a piezoelectric force microscope. A graphic processing method is taken to evaluate the domain retention. A domain life time of 25.0 h is obtained and possible mechanisms are discussed. Our study gives a direct reference for domain structure-related applications of LNOI, including guiding wave nonlinear frequency conversion, nonlinear wavefront tailoring, electro-optic modulation, and piezoelectric devices.

We observe the symmetrical dual-valley spectrum in periodically poled lithium niobate (PPLN) with a defect in the middle employing electro-optic effect. When the external electric fields are applied, light satisfying phase-matching condition (λ 0 ) will undergo polarization rotation and then turn back. However, the polarization states of wavelength at λ 0 ±Δλ will be changed at the end of the sample due to phase mismatching. A dual-valley transmission spectrum is thus obtained. Moreover, the depth of the valley could be tailored with different electric fields on the y-surface of the sample.

Optical-quality single-crystal LiB 3 O 5 (LBO) nanowires are synthesized for the first time using a sol–gel method. The LBO nanowires possess diameters ranging from 200 to 800 nm and lengths of up to 200 μm, and exhibit excellent uniformity, smooth surfaces, and good mechanical properties. A typical propagating loss of 0.038 dB/μm at 532 nm is obtained for a 620 nm-diameter nanowire. This is a decrease of one order of magnitude compared with that of a β-BaB 2 O 4 (BBO) nanowire with similar diameter, which makes the LBO nanowire a promising candidate to construct miniaturized nonlinear photonic devices.